NVIDIA GeForce 4XX Series Discussion

[Solaris17]

Ruby looks like a russian dyke in cosplay gear. I don't even remember what Nvidia uses as its mascot. I know Nvidia used some creepy model at one point though.

they use the marmaids primarily though during the 5 series they used some fugly ass fairy

EDIT: her their names were dawn and dusk sisters

 

[DaedalusHelios]

they use the marmaids primarily though during the 5 series they used some fugly ass fairy

EDIT: her their names were dawn and dusk sisters

http://static.desktopnexus.com/wallpapers/150733-bigthumbnail.jpg

Nah that was back in the FX5200,FX5900 days. Old stuff.

The most recent demo was medusa for gt200.

 

[Solaris17]

Nah that was back in the FX5200,FX5900 days. Old stuff.

The most recent demo was medusa for gt200.

http://img.techpowerup.org/090929/medusa.jpg

ya i knew they were 5 series...i just thought nthey were sticking with the mermaids......that medusa is rough though damn...make market are they trying to target?

 

[DaedalusHelios]

ya i knew they were 5 series...i just thought nthey were sticking with the mermaids......that medusa is rough though damn...make market are they trying to target?

Well they don't use a digital girl to help sell their product or branding anymore really. They just use the Nvidia emblem and statistics. I think they target an older crowd than ATi sometimes. Nvidia's adverts always have mid-twenties to early-thirties people "enjoying" the product. It makes sense as those people have more to spend, because they make more money(generally speaking).

 

[10TaTioN]

[Rumour] Nvidia GT300 Details - Fermi Codename Revealed

The codename is Fermi, after the renowned nuclear physicist, Enrico Fermi, the man behind Chicago Pile-1, the very first artificial nuclear reactor. While AMD have gone for green, Nvidia have gone explosive - perhaps indicative of their respective strategies.

It is clear that Nvidia are concentrating as much, if not more, on parallel computing rather than gaming. GT300 is expected to be a revolution, rather than the evolution Cypress was. GT300 will be large and hot, and is rumoured to be faster than the HD 5870. It is unknown as to how the GT300 will compete with Hemlock (5870 X2) - although Fudzilla suggest a dual-GT300 version is being planned. That could mean - a) A significantly cut down version of GT300 x 2; b) GT300 is more efficient than we are expecting; c) World's first widely distributed >300W TDP card. (Asus Mars is the only one thus far, though it is very much a limited edition) or d) Nvidia have to wait till the next half-node die shrink, like they did with GT200.

More @ Vr-Zone

 

[erocker]

Anyone have a look at the bottom of this chart yet? http://www.techarp.com/showarticle.aspx?artno=88&pgno=7

 

[Bo_Fox]

http://img.techpowerup.org/090929/physx.png


what?

HUH?? I read somewhere that Nvidia was ditching support for Ageia cards under Win7. Must've been a false rumor!

Just looked it up on google and it says that Nvidia disabled it whenever an ATI card is present. That means, if you have a 5870 and a dedicated Geforce card for Physx, it wouldnt work unless you use older PhysX drivers. http://www.xbitlabs.com/news/video/display/20090929095918_Nvidia_Reportedly_Limits_PhysX_Support_to_Nvidia_Only_Graphics_Sub_Systems.html

 

[Binge]

Anyone have a look at the bottom of this chart yet? http://www.techarp.com/showarticle.aspx?artno=88&pgno=7

Wait wait... what is that source supposed to represent? I mean are they worth giving credit for accuracy?

 

[Deleted member 67555]

Anyone have a look at the bottom of this chart yet? http://www.techarp.com/showarticle.aspx?artno=88&pgno=7

nice find Erocker

 

[erocker]

Wait wait... what is that source supposed to represent? I mean are they worth giving credit for accuracy?

As far as I know, they are pretty reputable. However, I have no idea where they got this information or if it's real.

 

[Solaris17]

As far as I know, they are pretty reputable. However, I have no idea where they got this information or if it's real.

we will know soon enough ill keep an eye on some of the drivers i can get from one of my buds at BFG...it may take a little bit of time...but im sure before long i can get the official names of the GT300 cards.

 

[qubit]

Nvidia to show DX 11 Fermi today

Fudzilla seems to have all the dirt on this lately.

Nvidia will show its DirectX 11 card that we used to call GT300 and that everyone internally calls Fermi. The hardware is real and we will learn much more about it tomorrow. We expect that Nvidia CEO Jensen Huang will demonstrate the card at the GTC keynote that should start at 1.00 PM Pacific time or roughly 10 PM Central European time.

 

[wolf]

Fudzilla seems to have all the dirt on this lately.

Nvidia will show its DirectX 11 card that we used to call GT300 and that everyone internally calls Fermi. The hardware is real and we will learn much more about it tomorrow. We expect that Nvidia CEO Jensen Huang will demonstrate the card at the GTC keynote that should start at 1.00 PM Pacific time or roughly 10 PM Central European time.

tomorrow they say? awesome

can't wait to see how she compares to cypress, and lets face it, hemlock.

 

[Benetanegia]

Anyone have a look at the bottom of this chart yet? http://www.techarp.com/showarticle.aspx?artno=88&pgno=7

If that is true, GT300 could fly and could destroy every already released card. By numbers:


Card ---------- Pixel FIll Rate ------- Texture FIll Rate ------ GFlops ----- Memory BandWidth

8800GT ------- 9,600 Mpixels/s ----- 33,600 Mtexels/s ----- 504.000 ---- 57.60 GB/s

9800GTX+ ---- 11,808 MPixels/s ---- 47,232 MTexels/s ----- 705.024 ---- 70.40 GB/s

GTX260 ------ 16,128 Mpixels/S ---- 36,864 Mtexels/s ----- 715.392 ---- 111.89 GB/s

GTX285 ------ 20,736 MPixels/s ---- 51,840 MTexels/s ----- 1062.720 --- 158.98 GB/s

GTS 340 ----- 25,000 MPixels/s ---- 50,000 Mtexels/s ------ 1440.000 --- 128.0 GB/s

GTX 380 ----- 44,800 MPixels/s ---- 89,600 Mtexels/s ------ 2457.600 --- 281.6 GB/s


If the jump to GTX3xx cards is anything closely as efficient to the jump to GTX2xx cards*, they could certainly be much faster.

*In the sense that no matter that the Texture is lower if the GFlops are higher.

Performance in Wizzards charts do seem to corroborate that overall performance is more related to GFlops than anything else. It's going to be interesting if they trully show them today.

 

[WhiteLotus]

If that information is true, then the GTX380 is going to be a monster at nearly 2.5 flops.

 

[qubit]

If that information is true, then the GTX380 is going to be a monster at nearly 2.5 flops.

I'll certainly be looking to get one when the price drops in a few months.

 

[DaedalusHelios]

I'll certainly be looking to get one when the price drops in a few months.

Yeah. I hope its going to be reasonably priced. I am not doubting whether it will be a great card. I am wondering what price it will command in a slightly weak global economy.

 

[Benetanegia]

If that information is true, then the GTX380 is going to be a monster at nearly 2.5 flops.

Tflops . Yeah I had it calculated in my post.

The most impresssive of that IMO is that the 0.7 Tflop GTX260 is as fast as the 1.2 Tflop Hd4870.

If the same relation is mantained the GTX380 could compete with a theoretical 4.3 Tflop Ati card.

Or conversely in order to compete with 2.7 Tflop HD5870 and Nvidia card would just need 1,575 Tflops. According to the info in TechArp, the flops for GT300 cards would be*:

GTS340 (320 SP)= 1.44 Tflop
GTS350 (384 SP)= 1.96 Tflop**
GTX360 (448 SP)= 1.88 Tflop**
GTX380 (512 SP)= 2.46 Tflop

That could mean, but not confirm in any case***, that Nvidia could have 3 GT300 cards above the HD5870. And that means a lot of competition.

*To calculate that I did two assumptions although based on leaked info:

1- Texture units and SPs are still linked in clusters like in G80/G92/GT200.
2- There are 32 SPs per cluster, instead of 24 in GT200 and 16 in G80/92.

Both have been leaked from more than one source. It also fits the leaked info for the GTX380, because if we divide Texture Fillrate by the clock frequency we obtain 128 texture units.

** Strange that the lower card has more flops, but it did happen with 8800GS/9600GSO and the 9600GT.
*** Its not that hard to believe anyway. Hd5870 is 30% faster than GTX285, wich has 240 SP. 320 already is 50% more, 384 is 60% more, so a lot of things can happen...

 

[Benetanegia]

Yeah. I hope its going to be reasonably priced. I am not doubting whether it will be a great card. I am wondering what price it will command in a slightly weak global economy.

Yeah, I'm wondering too. But if they are releasing 4 cards which on paper look better than any GT200 card, and GTS cards look cheaper to produce than GTX2xx cards, so even if they release the GTX380 very expensive, the other ones will be cheaper IMO:

GTX380 $500
GTX360 $400
GTS350 $300
GTS340 $250

That while having a very good margin on every card.

 

[DaedalusHelios]

Lets hope its:

GTX395 $600
GTX380 $400
GTX360 $300
GTS350 $250
GTS340 $200

 

[Benetanegia]

Lets hope its:

GTX395 $600
GTX380 $400
GTX360 $300
GTS350 $250
GTS340 $200

IMO they could do that easily and even lower if they trully wanted to own Ati instead of making big profits, but that is unlikely. Let's be honest the only difference in production cost between a GTX285 and the GTX380 will be the GPU, PCB is going to be similar, memory unless there's only a 2GB version the same, power circuitry is going to be similar if the rumors about a GTX395 are true. With GTX360 the same but with the GTX260 which is selling for $150, so production casts derived from PCB+memory+power circuitry aren't all that high.

Even according to the always taken out of proportion numbers of Charlie Demerjian GT300 chip would cost $100 to produce, while GT200 costs $50 according to him again. If they trully trully wanted to destroy AMD, they could release even this IMHO:

GTS340 $150
GTS350 $175
GTX360 $250
GTX380 $350

But it should be noted that AMD could also price their cards:

HD5850 $150
HD5890 $250

The thing is that IMO that low-profit situation would damage AMD much more than Nvidia, who could still make a lot of money from Tesla and Quadro. Both account for %50 of Nvidia's revenue. Not to mention Nvidia's capital is much bigger thanks to 8800 days. That scenario is very very very unlikely and at least at launch, what I first posted is what has a bigger chance of happening.

 

[btarunr]

Some specs
http://www.techpowerup.com/104942/NVIDIA_GT300__Fermi__Detailed.html

 

[[I.R.A]_FBi]

iSkeet

 

[Benetanegia]

Some specs
http://www.techpowerup.com/104942/NVIDIA_GT300__Fermi__Detailed.html

So which "source" should we believe?

This doesn't contradict all the other info, just the one regarding the memory and ROP count, which was sick and unnecesary, anyway. The rest is more like confirmed. Could it be that TechArp was only told that there were going to be 4 versions with every one of them having some disabled clusters and they started from the wrong place? Could it be 384, 320, 256, 192? Probably both GTS cards would share a 256 bit in that case though.

I'm just speculating, I like so.

In any case that would mean that on the PCB side GTX3xx is cheaper than GTX2xx. This is getting better by the moment.

Half Speed IEEE 754 Double Precision is just sick!!!

 

[[I.R.A]_FBi]

Half Speed IEEE 754 Double Precision is just sick!!!

What does that mean to the uninitiated like myself?